Rep Experiments (Deprecated but working)

Guidelines to run the experiments - see also moviekg

Overview

1. Prerequesits
2. Running Docker Only
3. Running

Prerequesits

• 32-64GB memory, 50GB disk space, (2GB nvidia GPU)
• git, make, tar, g(un)zip, docker, (uv)

Running Docker Only (recommended)

In this mode the KGpipe framework itself is running in docker and needed tasks that are using docker are called by it using the host docker sock as a mount.

Minimal example docker.env file for configuring the experiment in docker, like locations.

Copy docker_env to docker.env when finished

# .env
PIPELINE_CONFIG=pipeline.conf

DATASET_SELECT=small

# path inside docker
ONTOLOGY_PATH=/app/experiments/moviekg/movie-ontology.ttl

EMBEDDER=sentence-transformer
EMBED_CACHE="redis://cache:6379"

DBPEDIA_ANNOTATE_URL='http://dbpedia-spotlight:80/rest/annotate'
OPENAI_TOKEN="INSERT_YOUR_OPENAI_TOKEN_HERE"
DEFAULT_LLM_MODEL_NAME=gpt-5-mini

# OPTIONAL
OLLAMA_TOKEN="empty"
LLM_ENDPOINT_URL=https://example.com/ollama/api/generate

Prepare

make setup_docker

Execution of dataset stats, pipelines, evalaution, and paper content generation

make run_docker_small

Running (WIP)

1. Pipelines
2. Evaluation
3. Paper Content

Pipelines

To execute all pipelines run

make pipelines

Outputs are in $OUTPUT/${pipeline_name}

Evaluation

To create statistics for

make evaluation

Outputs are in $OUTPUT/${pipeline_name}_metrics.csv and aggregated metrics in $OUTPUT/all_metrics.csv

Paper Content

Creates the figures and tables for the papers. Not in latex format therefore I recommend using https://www.latex-tables.com/

make paper

Outputs are in $OUTPUT/paper